Dual-Branch Discrimination Network Using Multiple Sparse Priors for Image Deblurring.

Blind image deblurring is a challenging problem in computer vision, aiming to restore the sharp image from blurred observation. Due to the incompatibility between the complex unknown degradation and the simple synthetic model, directly training a deep convolutional neural network (CNN) usually cannot sufficiently handle real-world blurry images. An existed generative adversarial network (GAN) can generate more detailed and realistic images, but the game between generator and discriminator is unbalancing, which leads to the training parameters not being able to converge to the ideal Nash equilibrium points. In this paper, we propose a GAN with a dual-branch discriminator using multiple sparse priors for image deblurring (DBSGAN) to overcome this limitation. By adding the multiple sparse priors into the other branch of the discriminator, the task of the discriminator is more complex. It can balance the game between the generator and the discriminator. Extensive experimental results on both synthetic and real-world blurry image datasets demonstrate the superior performance of our method over the state of the art in terms of quantitative metrics and visual quality. Especially for the GOPRO dataset, the averaged PSNR improves 1.7% over others.

Quantifying the performance of MEG source reconstruction using resting state data.

In magnetoencephalography (MEG) research there are a variety of inversion methods to transform sensor data into estimates of brain activity. Each new inversion scheme is generally justified against a specific simulated or task scenario. The choice of this scenario will however have a large impact on how well the scheme performs. We describe a method with minimal selection bias to quantify algorithm performance using human resting state data. These recordings provide a generic, heterogeneous, and plentiful functional substrate against which to test different MEG recording and reconstruction approaches. We used a Hidden Markov model to spatio-temporally partition data into self-similar dynamic states. To test the anatomical precision that could be achieved, we then inverted these data onto libraries of systematically distorted subject-specific cortical meshes and compared the quality of the fit using cross validation and a Free energy metric. This revealed which inversion scheme was able to identify the least distorted (most accurate) anatomical models, and allowed us to quantify an upper bound on the mean anatomical distortion accordingly. We used two resting state datasets, one recorded with head-casts and one without. In the head-cast data, the Empirical Bayesian Beamformer (EBB) algorithm showed the best mean anatomical discrimination (3.7 mm) compared with Minimum Norm/LORETA (6.0 mm) and Multiple Sparse Priors (9.4 mm). This pattern was replicated in the second (conventional dataset) although with a marginally poorer (non-significant) prediction of the missing (cross-validated) data. Our findings suggest that the abundant resting state data now commonly available could be used to refine and validate MEG source reconstruction methods and/or recording paradigms.

Algorithmic procedures for Bayesian MEG/EEG source reconstruction in SPM.

The MEG/EEG inverse problem is ill-posed, giving different source reconstructions depending on the initial assumption sets. Parametric Empirical Bayes allows one to implement most popular MEG/EEG inversion schemes (Minimum Norm, LORETA, etc.) within the same generic Bayesian framework. It also provides a cost-function in terms of the variational Free energy-an approximation to the marginal likelihood or evidence of the solution. In this manuscript, we revisit the algorithm for MEG/EEG source reconstruction with a view to providing a didactic and practical guide. The aim is to promote and help standardise the development and consolidation of other schemes within the same framework. We describe the implementation in the Statistical Parametric Mapping (SPM) software package, carefully explaining each of its stages with the help of a simple simulated data example. We focus on the Multiple Sparse Priors (MSP) model, which we compare with the well-known Minimum Norm and LORETA models, using the negative variational Free energy for model comparison. The manuscript is accompanied by Matlab scripts to allow the reader to test and explore the underlying algorithm.

Multiple sparse volumetric priors for distributed EEG source reconstruction.

We revisit the multiple sparse priors (MSP) algorithm implemented in the statistical parametric mapping software (SPM) for distributed EEG source reconstruction (Friston et al., 2008). In the present implementation, multiple cortical patches are introduced as source priors based on a dipole source space restricted to a cortical surface mesh. In this note, we present a technique to construct volumetric cortical regions to introduce as source priors by restricting the dipole source space to a segmented gray matter layer and using a region growing approach. This extension allows to reconstruct brain structures besides the cortical surface and facilitates the use of more realistic volumetric head models including more layers, such as cerebrospinal fluid (CSF), compared to the standard 3-layered scalp-skull-brain head models. We illustrated the technique with ERP data and anatomical MR images in 12 subjects. Based on the segmented gray matter for each of the subjects, cortical regions were created and introduced as source priors for MSP-inversion assuming two types of head models. The standard 3-layered scalp-skull-brain head models and extended 4-layered head models including CSF. We compared these models with the current implementation by assessing the free energy corresponding with each of the reconstructions using Bayesian model selection for group studies. Strong evidence was found in favor of the volumetric MSP approach compared to the MSP approach based on cortical patches for both types of head models. Overall, the strongest evidence was found in favor of the volumetric MSP reconstructions based on the extended head models including CSF. These results were verified by comparing the reconstructed activity. The use of volumetric cortical regions as source priors is a useful complement to the present implementation as it allows to introduce more complex head models and volumetric source priors in future studies.

Source localization of epileptic spikes using Multiple Sparse Priors.

OBJECTIVE: To evaluate epileptic source estimation using multiple sparse priors (MSP) inverse method and high-resolution, individual electrical head models. METHODS: Accurate source localization is dependent on accurate electrical head models and appropriate inverse solvers. Using high-resolution, individual electrical head models in fifteen epilepsy patients, with surgical resection and clinical outcome as criteria for accuracy, performance of MSP method was compared against standardized low-resolution brain electromagnetic tomography (sLORETA) and coherent maximum entropy on the mean (cMEM) methods. RESULTS: The MSP method performed similarly to the sLORETA method and slightly better than the cMEM method in terms of success rate. The MSP and cMEM methods were more focal than sLORETA with the advantage of not requiring an arbitrary selection of a hyperparameter or thresholding of reconstructed current density values to determine focus. MSP and cMEM methods were better than sLORETA in terms of spatial dispersion. CONCLUSIONS: Results suggest that the three methods are complementary and could be used together. In practice, the MSP method will be easier to use and interpret compared to sLORETA, and slightly more accurate and faster than the cMEM method. SIGNIFICANCE: Source localization of interictal spikes from dense-array electroencephalography data has been shown to be a reliable marker of epileptic foci and useful for pre-surgical planning. The advantages of MSP make it a useful complement to other inverse solvers in clinical practice.

Comparison of Machine Learning Techniques Based Brain Source Localization Using EEG Signals.

BACKGROUND: The brain is the most complex organ of the human body with millions of connections and activations. The electromagnetic signals are generated inside the brain due to a mental or physical task performed. These signals excite a bunch of neurons within a particular lobe depending upon the nature of the task performed. To localize this activity, certain machine learning (ML) techniques in conjunction with a neuroimaging technique (M/EEG, fMRI, PET) are developed. Different ML techniques are provided in the literature for brain source localization. Among them, the most common are: minimum norm estimation (MNE), low resolution brain electromagnetic tomography (LORETA) and Bayesian framework based multiple sparse priors (MSP). AIMS: In this research work, EEG is used as a neuroimaging technique. METHODS: EEG data is synthetically generated at SNR=5dB. Afterwards, ML techniques are applied to estimate the active sources. Each dataset is run for multiple trials (>40). The performance is analyzed using free energy and localization error as performance indicators. Furthermore, MSP is applied with a variant number of patches to observe the impact of patches on source localization. RESULTS: It is observed that with an increased number of patches, the sources are localized with more precision and accuracy as expressed in terms of free energy and localization error, respectively. CONCLUSION: The patches optimization within the Bayesian Framework produces improved results in terms of free energy and localization error.

Bayesian Algorithm Based Localization of EEG Recorded Electromagnetic Brain Activity.

BACKGROUND: Electrical signals are generated inside human brain due to any mental or physical task. This causes activation of several sources inside brain which are localized using various optimization algorithms. METHODS: Such activity is recorded through various neuroimaging techniques like fMRI, EEG, MEG etc. EEG signals based localization is termed as EEG source localization. The source localization problem is defined by two complementary problems; the forward problem and the inverse problem. The forward problem involves the modeling how the electromagnetic sources cause measurement in sensor space, while the inverse problem refers to the estimation of the sources (causes) from observed data (consequences). Usually, this inverse problem is ill-posed. In other words, there are many solutions to the inverse problem that explains the same data. This ill-posed problem can be finessed by using prior information within a Bayesian framework. This research work discusses source reconstruction for EEG data using a Bayesian framework. In particular, MSP, LORETA and MNE are compared. RESULTS: The results are compared in terms of variational free energy approximation to model evidence and in terms of variance accounted for in the sensor space. The results are taken for real time EEG data and synthetically generated EEG data at an SNR level of 10dB. CONCLUSION: In brief, it was seen that MSP has the highest evidence and lowest localization error when compared to classical models. Furthermore, the plausibility and consistency of the source reconstruction speaks to the ability of MSP technique to localize active brain sources.